Skip to main content
These examples use the direct API host, https://api.nano-gpt.com. It has no IPv6 address, so clients with only IPv6 should use https://nano-gpt.com with the same paths. See API Hosts.

​
SillyTavern Integration

SillyTavern has NanoGPT built in, so you only need your API key. Use the NanoGPT source rather than the OpenAI or Custom source: it knows the model list, reasoning and NanoGPT’s extra samplers.
Tested setup. Last verified November 6, 2026 with SillyTavern 1.19.0 and DeepSeek V4.1 Flash 1017 Thinking. We ran the steps below in a real SillyTavern, then checked a streamed reply, reasoning, a lorebook entry, a reply cut off partway, and Continue. Requests went to NanoGPT through a local test proxy, which removed SillyTavern’s neutral sampler values before sending them.

​
Set up NanoGPT

  1. Open API Connections (the plug icon).
  2. Set API to Chat Completion.
  3. Set Chat Completion Source to NanoGPT.
  4. Paste your key into NanoGPT API Key and click Connect. Create a key at nano-gpt.com/api.
  5. Pick a model under NanoGPT Model, for example openai/gpt-6.1-sol. The list comes from NanoGPT.
SillyTavern’s NanoGPT source always connects to https://nano-gpt.com/api/v1, so there is no address to enter.

​
Settings that matter

  • Streaming: leave it on. If a reply fails partway, you keep what already arrived.
  • Response (tokens): the default is 300. Thinking models spend part of this on reasoning, so raise it for longer replies, or the reply can end early.
  • Top K, Top A, Min P, Repetition Penalty: leave them at their defaults (0, 0, 0 and 1) unless you know the model uses them. Some models, including the uncensored ones, have no route that accepts them. NanoGPT then returns an error that names the setting: put that setting back to its default.

​
Lorebooks and character cards

SillyTavern handles World Info, lorebooks and character cards itself and adds the matching text to each request. They need no NanoGPT setting. In our test, an entry triggered by a keyword reached the model, and the model answered from it.

​
If a reply stops partway

SillyTavern shows an error and keeps the text it received. Click Regenerate for a fresh reply, or Continue to carry on from the partial text.
  • Continue can repeat the last few words, because the model starts from the cut-off text. It repeated them in our test, so check the join.
  • Text that reached you before the failure is normally charged. Regenerate or Continue is a new request.
  • If a reply just ends with no error, SillyTavern can’t tell it was cut off. Use the same two buttons.

​
Image Generation

SillyTavern has a built-in NanoGPT image source, so you do not need to enter an image API URL.
  1. Save your NanoGPT API key under API Connections (Chat Completion → NanoGPT). The image source reuses that key.
  2. Open Extensions → Image Generation and set Source to NanoGPT.
  3. Pick a model from the Model list. SillyTavern loads it from NanoGPT’s image catalog.
  4. Choose a Resolution the model supports. SillyTavern defaults to 512x512, which some models reject. GPT Image 2.5, for example, needs at least 1024x1024 (or another supported size such as 1536x1024).
SillyTavern’s OpenAI image source always calls OpenAI directly and cannot be pointed at NanoGPT. Use the NanoGPT source instead.
SillyTavern reads images from the inline b64_json field. Very large outputs (roughly over 3.5 MB, for example 4K PNGs) are returned as a download url instead, which SillyTavern cannot display. Use a smaller resolution if generation fails for large sizes.